TSMC and an AI Bubble
The speaker discusses historical parallels between AI investment and past tech bubbles, noting key differences that may prevent a severe crash. TSMC's control over wafer supply is identified as a critical constraint that could naturally limit overbuilding. The speaker argues that if TSMC resists pressure to massively expand capacity, it may single-handedly prevent an AI bubble.
Summary
The speaker opens by acknowledging that historical precedent for foundational technologies strongly suggests an AI bubble is likely, as bubbles have typically accompanied the buildout of transformative technologies. These bubbles tend to fund infrastructure development but often result in supply outpacing demand, leading to a market crash — especially severe when debt-fueled, as seen in the dot-com crash of 2000.
However, the speaker highlights several key differences between the current AI buildout and the year 2000 bubble. Most notably, the current buildout is predominantly funded through operating cash flows rather than debt, which significantly reduces systemic financial risk. Additionally, unlike the dot-com era where 99% of fiber optic infrastructure sat unutilized, today's GPUs are running at 100% utilization, indicating genuine demand rather than speculative overbuilding.
The speaker then focuses on TSMC's role as a critical bottleneck in the semiconductor supply chain. Because TSMC controls the production of advanced wafers, it effectively acts as a natural cap on how many GPUs Nvidia can bring to market. The speaker estimates that if TSMC fully accommodated Nvidia CEO Jensen Huang's ambitions, Nvidia could potentially sell $2–3 trillion worth of GPUs in 2026 or 2027, which could trigger a consumption-driven overbuild and eventual bubble. By maintaining supply constraints, TSMC may inadvertently — or deliberately — be preventing that outcome, leading the speaker to humorously suggest that TSMC deserves a 'giant party' in Taiwan for potentially saving the market from a bubble.
Key Insights
- The speaker argues that every foundational technology in history has produced a bubble, and AI should be no exception — but the severity depends heavily on whether the buildout is debt-fueled, as in 2000, versus cash-flow funded, as it largely is today.
- The speaker contrasts current GPU utilization (near 100%) with the dot-com era's 99% unutilized fiber optic infrastructure, arguing this real demand fundamentally differentiates the current AI cycle from speculative past bubbles.
- The speaker claims that TSMC's control over advanced wafer supply is currently the primary structural constraint preventing Nvidia — and by extension the AI industry — from overbuilding into bubble territory.
- The speaker estimates that if TSMC fully met Jensen Huang's demand, Nvidia could sell $2–3 trillion worth of GPUs in 2026 or 2027, a scale the speaker believes would likely result in a consumption overbuild and subsequent market crash.
- The speaker credits TSMC with potentially single-handedly preventing an AI bubble by acting as a supply bottleneck, humorously suggesting Taiwan deserves a 'giant party' if the bubble is avoided.
Topics
Transcript
[0:00] We should expect a bubble. Every prior market precedent for a foundational new technology like AI, you've always had a bubble. That bubble funds the buildout of this new technology, but supply gets ahead of demand and you get a crash and it's a particularly severe crash if it's a debtfueled buildout like the year 2000. And one thing really good about the current buildout is it's still overwhelmingly funded out of operating cash flows, which is a a really important fundamental difference versus the year 2000. has is valuation has is [0:30] the fact that every GPU is running at 100% utilization when 99% of fiber was unutilized. So there's all these fundamental differences and I have been…
Full transcript available for MurmurCast members
Sign Up to AccessMore from Invest Like The Best
How Losing Everything Shaped Uber’s CEO
Uber's CEO reflects on his immigrant experience, describing how his family lost everything after coming from Iran and how that hardship shaped his drive to rebuild. He credits life's challenges as formative experiences that provide profound human satisfaction when overcome. He emphasizes a stable, grounded sense of self as his anchor against external chaos.
Why Uber's Future is Autonomous
Uber is positioning itself as the dominant aggregator of autonomous vehicle supply by partnering with major AV providers. The company is building ecosystem infrastructure and collecting real-world street data to accelerate AV deployment and capture instant demand when vehicles hit new markets.
Why the AI Boom Is Just Getting Started
Investor Alex from Whale Rock Capital discusses their high-conviction investment in Anthropic, explaining why AI represents an unprecedented 'L-curve' of adoption rather than a typical S-curve. He outlines their framework of S-curves, competitive advantages, and underappreciated earnings power, while detailing the decommoditization of hardware infrastructure and concerns about traditional enterprise software companies.
Uber CEO on AI, Autonomous Vehicles, and the Future of Transportation
Uber CEO Dara Khosrowshahi discusses his journey from Expedia to leading Uber through chaos, the company's AI and autonomous vehicle strategy, and his personal philosophy on leadership and resilience. He covers Uber's expansion into hotels, membership programs, and the physical AI opportunity, while reflecting on mentors like Barry Diller and the lessons from his immigrant upbringing.
30 Years of Finding Alpha | Dan Loeb
Dan Loeb, founder of Third Point, reflects on 30 years of investing across credit, equities, and activism, discussing his evolution from event-driven deep value to quality growth investing. He shares views on AI's transformative impact on markets and portfolio construction, governance lessons from activism campaigns, and insights on the future role of human capital allocators in an increasingly AI-driven world.